Looking at Magnitude and Phase in the STFT

In this part of the assignment we will modify time and pitch independently using the stft, which is historically often refered to as a "phase vocoder" in audio processing.

from a3py import *
import scipy as sp

def newfig(title=None):
    fig = plt.figure(figsize=(12,3));
    if(title is not None):
        print(title)
        fig.suptitle(title, fontsize=16);

Let's grab some audio...

audio_file_path = './noisyhellocs6682.wav';
a = Audio(audio_file_path);
print("Input Audio File {}:".format(audio_file_path));
a.play()
stft = STFT.FromAudio(a);
stft.show()

Input Audio File ./noisyhellocs6682.wav:

Which is more important: Phase or Magnitude?

Here we will do an experiment where we compare what happens when we randomize the phases of an stft and take its istft to what happens if we randomize the magnitudes.

# You should finish the implementation of this part

Let's see what happens when we use a relatively small window size and hop size...

window_size=1024;
hop_size = 128;
compareRandPhaseMag(a, window_size=window_size, hop_size=hop_size)

Unmodified ISTFT:

Original Phases, Random Magnitudes ISTFT:

Original Magnitudes, Random Phases ISTFT:

So, which is more important in the short time domain? Phase or Magnitude? How does this depend on window size and hop size? Why might window size and hop size matter?

Write your answer below in this markdown cell... you don't need to write more than a few sentences for each.

How do smaller/larger window and hop sizes affect the importance of phase and magnitude?:

[Your answer here]

Hint: Remember our discussion of randomizing phase vs magnitude in the non-short time Fourier domain. What hop size and window size does this correspond to?